Anthropic pitches draft federal AI safety bill to lawmakers

The pitch was made by Anthropic representatives to Capitol Hill lawmakers, and the proposal is a draft bill on frontier AI safety and security.[1] POLITICO said it obtained the confidential draft, which had not previously been reported, and based its reporting on accounts from three people with knowledge of the private deliberations, granted anonymity to discuss them.[1] The proposal concerns when the government could block the release or use of an advanced AI model, as well as the reach of state AI safety laws.[1] The available evidence describes a pitch and a draft, not an enacted federal law.[1]

Under the draft, the government could block a model’s release or use only if it posed a foreseeable, imminent risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people.[1] The specified risk sources include creating or releasing a biological weapon, a model evading its developer’s controls, and a cyberattack.[1] These are conditions in the proposed blocking standard, not a report that any model has triggered them.[1] For teams assessing the proposal, the stated figures and the requirement that risk be foreseeable and imminent are central details to track.[1]

The draft would preempt some state AI safety laws, and POLITICO reported that its language mirrors language in California AI laws, including SB 53.[1] POLITICO reported that California SB 53 was signed into law in 2025.[1] Anthropic’s June 2026 Advanced AI Framework takes a conditional position on federal preemption: it says Congress should preempt state law only if it enacts a rigorous federal regime that meets or exceeds the framework’s strongest proposed measures for catastrophic AI risks.[2] The framework also says that, without a strong federal law, states should retain full authority to legislate.[2] The evidence does not establish whether the draft meets that condition.[1][2]

Anthropic says the bill represents proposals in its Advanced AI Framework and that it has advocated for those proposals with federal and state lawmakers since June.[1] The framework, dated June 2026, sets out near-term recommendations for governments on serious frontier AI risks.[2] Its proposals include testing the most capable models for catastrophic risks, engaging external evaluators, disclosing results, risks, and incidents on an ongoing basis, and being accountable to government.[2] It also proposes cross-government and cross-sector investments intended to make biological or cyberattacks harder to carry out and easier to recover from.[2] Anthropic says it considers these policies sensible and believes Congress should act on them.[1]